Rapid detection method for traumatic brain injury fusing gene engineering and quantum dots

Abstract

The application discloses a rapid detection method for traumatic brain injury by combining gene engineering and quantum dots, and relates to the technical field of traumatic brain injury detection. The method obtains a biological sample of a suspected patient, extracts biomolecular components, compares the biomolecular components with a biomolecular database of healthy people, and obtains relevant differentially expressed molecules. A recombinant antibody capable of specific recognition is constructed by gene engineering, and is connected with quantum dots through a coupling reaction to form a detection probe. The probe is mixed with the biological sample for a reaction, and a fluorescence signal intensity generated by combination is collected by a fluorescence detection device. After noise reduction and feature extraction by a signal processing algorithm, a standardized signal value is obtained, a sample is evaluated in combination with a bioinformatics database, and a preliminary detection result is determined. Finally, according to the differentially expressed molecules, the standardized signal value and the preliminary result, a multivariate statistical model is used to analyze a data change trend in a preset time window, and a diagnosis report is generated. The method combines multiple technologies and provides a new path for traumatic brain injury detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traumatic brain injury detection, in particular to a rapid detection method of traumatic brain injury by combining genetic engineering and quantum dots. BACKGROUND

[0002] Traumatic brain injury is a common neurological emergency caused by external factors such as traffic accidents, falls, and impacts. The degree of injury is closely related to the prognosis, and timely and accurate diagnosis is of great significance for the selection of subsequent treatment options. Currently, there are various methods for detecting traumatic brain injury in clinical practice, among which imaging examinations such as computed tomography and magnetic resonance imaging can clearly show structural damage to brain tissue, including hemorrhage, edema, and contusion. However, these examinations rely on large and precise equipment, have complex operation procedures, and require professional personnel for operation and interpretation. In emergency rescue, field rescue, and other scenarios, it is difficult to achieve rapid deployment and immediate detection, often delaying valuable treatment time. Moreover, for some mild traumatic brain injuries, early structural changes in brain tissue are not obvious, and imaging examinations are prone to missed diagnosis.

[0003] Biomarker detection is another important detection method, which detects specific molecules such as proteins and nucleic acids related to traumatic brain injury in biological samples such as blood and cerebrospinal fluid to determine the occurrence and extent of injury. However, existing biomarker detection methods have many shortcomings. In terms of antibody acquisition, traditional antibody preparation relies on extraction after animal immunization, which not only has a long preparation period, but also has large batch-to-batch differences in antibody specificity and affinity due to animal individual differences and immunization processes, which can interfere with the accuracy of the detection results. At the same time, a large number of other biological molecules in the sample can non-specifically bind to the antibody, further affecting the specificity of the detection. In terms of signal detection, commonly used methods such as enzyme-linked immunosorbent assay have weak signal strength and are easily affected by environmental factors such as temperature and humidity, resulting in poor stability of the detection signal and low repeatability of the detection results.

[0004] Traumatic brain injury is a complex pathological process involving dynamic changes in multiple biological molecules, and detection of a single biomarker cannot fully reflect the overall picture and development trend of the injury. If multiple biomarkers are detected simultaneously, existing detection methods often require complex operation steps and long detection times, which cannot meet the clinical demand for rapid diagnosis. In actual clinical applications, due to the limitations of detection methods, there are often delays or misdiagnoses, affecting the treatment effect and prognosis of patients. SUMMARY

[0005] The present application aims to provide a rapid detection method of traumatic brain injury by combining genetic engineering and quantum dots to solve the problems raised in the background.

[0006] To achieve the above object, the application provides a rapid detection method for traumatic brain injury by combining gene engineering and quantum dots, which comprises the following steps:

[0007] The fluorescence signal intensity generated after the quantum dot-labeled detection probe is combined with the differentially expressed molecules is collected by a fluorescence detection device;

[0008] The fluorescence signal intensity is denoised and feature extracted by using a signal processing algorithm to obtain a standardized signal value, and the biological sample is evaluated based on the comparison result of the standardized signal value and a preset threshold value, and a preliminary detection result of traumatic brain injury is determined in combination with a bioinformatics database;

[0009] According to the differentially expressed molecules, the standardized signal value and the preliminary detection result, a multivariate statistical model is used to analyze the detection data change trend in a preset time window to generate a diagnosis report of traumatic brain injury.

[0010] Preferably, the fluorescence signal intensity is denoised and feature extracted by using a signal processing algorithm to obtain a standardized signal value, and the biological sample is evaluated based on the comparison result of the standardized signal value and a preset threshold value, and a preliminary detection result of traumatic brain injury is determined in combination with a bioinformatics database, which comprises the following steps:

[0011] The fluorescence signal intensity is multiscale decomposed by using a wavelet transform algorithm to remove noise interference components and obtain a denoised signal, and the denoised signal is feature dimension-reduced by using a principal component analysis method to extract key signal features;

[0012] The peak intensity, half-peak width and integral area of the fluorescence signal are calculated by using the key signal features, and the peak intensity, half-peak width and integral area are compared with a preset threshold value in the bioinformatics database to obtain a signal deviation degree;

[0013] The biological sample is classified by using a support vector machine classification model according to the signal deviation degree, a classification result is output, the classification result is optimized by using a cross-validation method, and a preliminary detection result of traumatic brain injury is obtained;

[0014] The detection reliability of the biological sample is verified by using the preliminary detection result in combination with a clinical symptom description, a verification report is generated, the preset threshold value is adjusted based on the verification report, and the preliminary detection result is updated.

[0015] Preferably, the biological sample is classified by using a support vector machine classification model according to the signal deviation degree, a classification result is output, the classification result is optimized by using a cross-validation method, and a preliminary detection result of traumatic brain injury is obtained, which comprises the following steps:

[0016] Optimizing the kernel function parameters of the support vector machine classification model by using a grid search method, determining the optimal parameter combination, inputting the signal deviation degree into the support vector machine classification model, and generating a preliminary classification result;

[0017] Based on the preliminary classification result, the performance of the support vector machine classification model is evaluated by using a K-fold cross-validation method, and the accuracy, sensitivity and specificity indexes are calculated;

[0018] According to the accuracy, sensitivity and specificity indexes, the decision boundary of the support vector machine classification model is adjusted, and the signal deviation degree is reclassified to obtain an optimized classification result;

[0019] The preliminary classification result, the accuracy, sensitivity and specificity indexes, and the optimized classification result are integrated to generate a preliminary detection result of traumatic brain injury.

[0020] Preferably, according to the accuracy, sensitivity and specificity indexes, the decision boundary of the support vector machine classification model is adjusted, and the signal deviation degree is reclassified to obtain an optimized classification result, comprising:

[0021] An evaluation function is constructed using the accuracy, sensitivity and specificity indexes, and the decision boundary parameters of the support vector machine classification model are adjusted by gradient descent method to maximize the evaluation function;

[0022] Based on the adjusted decision boundary parameters, the signal deviation degree is reclassified to generate an intermediate classification result;

[0023] The intermediate classification result and the preliminary classification result are subjected to consistency test, the Kappa coefficient is calculated, and the classification consistency is judged according to the Kappa coefficient;

[0024] If the Kappa coefficient is greater than a preset threshold, the intermediate classification result is taken as the optimized classification result; if the Kappa coefficient is less than or equal to the preset threshold, the decision boundary parameters are repeatedly adjusted and reclassified until the optimized classification result is obtained.

[0025] Preferably, the detection data change trend in the preset time window is analyzed by using a multivariate statistical model to generate a diagnosis report of traumatic brain injury, comprising:

[0026] A time series data matrix is constructed based on the differentially expressed molecules and standardized signal values, and the matrix is divided by a sliding window;

[0027] The mean, variance and autocorrelation coefficient in each time window are extracted as dynamic feature vectors;

[0028] inputting the dynamic feature vector into a pre-trained long short-term memory network model to output a trend prediction value;

[0029] generating a trend analysis atlas according to a deviation amount of the trend prediction value from historical detection data and integrating the trend analysis atlas into a diagnosis report.

[0030] Preferably, the inputting the dynamic feature vector into the pre-trained long short-term memory network model comprises:

[0031] updating a hidden layer weight of the long short-term memory network model using a time reversal propagation algorithm;

[0032] controlling a forgetting and remembering proportion of the feature vector through a gating mechanism;

[0033] regularizing a full connection layer using a dropout layer to output a normalized trend prediction value.

[0034] Preferably, the adjusting the preset threshold based on the verification report comprises:

[0035] calculating a moving average value of normalized signal values in five consecutive detections;

[0036] dynamically adjusting a floating range of the preset threshold according to a matching degree of the moving average value and a verification report result;

[0037] when the matching degree is lower than a set standard, triggering a re-calibration process of a quantum dot labeled detection probe.

[0038] Preferably, the triggering the re-calibration process of the quantum dot labeled detection probe comprises:

[0039] obtaining a standard distribution curve of historical fluorescence signal intensity, and performing outlier detection on current fluorescence signal intensity using a K-means clustering algorithm;

[0040] if a proportion of outliers exceeds an alarm value, starting a coupling reaction parameter optimization program of a recombinant antibody and a quantum dot.

[0041] Preferably, after the generating the preliminary detection result of the traumatic brain injury, the method further comprises:

[0042] establishing a multi-round iteration record library of a support vector machine classification model;

[0043] extracting a decision boundary parameter variation amount and a classification result difference of each iteration;

[0044] generating a model stability evaluation index by recognizing a parameter variation pattern through a convolutional neural network.

[0045] Preferably, the method further comprises:

[0046] The evaluation index is spatiotemporally correlated with signal deviation, confidence levels of detection data are divided according to a mapping result, and a final detection report framework is generated by fusing the confidence levels and an optimized classification result.

[0047] Compared with the prior art, the method has the following beneficial effects:

[0048] The recombinant antibody constructed by genetic engineering technology can accurately recognize the differentially expressed molecules related to traumatic brain injury, avoids the problem of insufficient specificity of natural antibodies, and reduces the interference of irrelevant components in the sample.

[0049] The fluorescence detection device collects the fluorescence signal, and the noise reduction and feature extraction of the signal processing algorithm can extract effective information from the complex signal background, so that the obtained standardized signal value can more truly reflect the existence state of the target molecule.

[0050] By comparing the standardized signal value with the preset threshold value and combining the evaluation of bioinformatics database, a preliminary detection result can be formed, which integrates multiple sources of information rather than relying on a single indicator, and can more comprehensively reflect the potential characteristics of traumatic brain injury.

[0051] The entire detection process from biological sample processing to diagnosis report generation is closely linked, the specificity of the recombinant antibody and the optical advantage of the quantum dots are used to improve the detection specificity, and the signal processing and multivariate analysis optimize the interpretation of the detection results from different aspects, so that the method can balance rapid response and comprehensive evaluation in the detection of traumatic brain injury, and meet the needs of clinical detection efficiency and information comprehensiveness. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The method flowchart of the fusion gene engineering and quantum dot rapid detection method of traumatic brain injury described in the application;

[0053] Figure 2 The flowchart of support vector machine classification optimization;

[0054] Figure 3 The flowchart of decision boundary adjustment and consistency test;

[0055] Figure 4 Flowchart for time series analysis and diagnostic report generation;

[0056] Figure 5 Flowchart for LSTM model training and prediction. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0058] Please refer to Figure 1 The present application provides a rapid detection method for traumatic brain injury by combining gene engineering and quantum dots, which comprises the following steps:

[0059] S1, acquiring the fluorescence signal intensity generated after the quantum dot labeled detection probe is combined with the differentially expressed molecules by a fluorescence detection device;

[0060] S2, using a signal processing algorithm to denoise and feature extract the fluorescence signal intensity, obtaining a standardized signal value, based on the comparison result of the standardized signal value and a preset threshold, combining a bioinformatics database to evaluate the biological sample, and determining a preliminary detection result of traumatic brain injury;

[0061] S3, according to the differentially expressed molecules, the standardized signal value and the preliminary detection result, using a multivariate statistical model to analyze the detection data change trend in a preset time window, and generating a diagnostic report of traumatic brain injury.

[0062] The implementation process of the method will be described in detail below.

[0063] A biological sample from a patient suspected of having traumatic brain injury is obtained, which includes blood, cerebrospinal fluid or saliva, etc. Biological molecule components in the biological sample are extracted by centrifugation, chromatography and other methods, which include proteins, nucleic acids, lipids, etc. The extracted biological molecule components are compared with a pre-established healthy population biological molecule database, and differential analysis software is used to screen out molecules in the biological molecule components whose expression amount deviates significantly from the reference range of healthy population, i.e. differentially expressed molecules that may be related to traumatic brain injury.

[0064] According to the amino acid sequence or nucleotide sequence of the differentially expressed molecules, design specific primers, which need to determine the complementary base sequence according to the amino acid or nucleotide sequence of the differentially expressed molecules, ensure accurate binding to the target gene fragment, lay the foundation for subsequent gene cloning of antibody fragment gene insertion into expression vectors, and ensure the specificity of recombinant antibodies. By gene cloning technology, the gene encoding the antibody fragment recognizing the differentially expressed molecules is inserted into the expression vector, transformed into the host cell for induction expression, and the recombinant antibody is purified by affinity chromatography. The recombinant antibody is mixed with quantum dot material at a certain molar ratio, a coupling agent (such as carbodiimide) is added, and the reaction is carried out at 25-37°C for 1-3 hours. The unbound quantum dots and coupling agent are removed by dialysis to prepare a quantum dot-labeled detection probe.

[0065] The quantum dot-labeled detection probe is mixed with the biological sample at a volume ratio of 1:5-1:10, and incubated at 37°C for 30-60 minutes to allow the detection probe to fully bind to the differentially expressed molecules in the biological sample. Using a fluorescence microscope or a fluorescence spectrophotometer and other fluorescence detection devices, the fluorescence signal intensity generated after binding is collected at a specific excitation wavelength, and the wavelength, intensity and duration of the fluorescence signal are recorded.

[0066] The collected fluorescence signal intensity is processed using a signal processing algorithm, first removing noise interference in the signal, then extracting key features and converting them into standardized signal values. The standardized signal values are compared with the preset threshold values, and relevant data in the bioinformatics database (such as GeneBank, UniProt, etc.) are called simultaneously, combined with the biological function information of the differentially expressed molecules, and the biological sample is comprehensively evaluated to determine the preliminary detection result of traumatic brain injury.

[0067] Collect the types and contents of differentially expressed molecules, changes in standardized signal values, and preliminary detection results, and arrange these data in chronological order. A multivariate statistical model is used to analyze the trend of detection data within a preset time window (such as 24 hours, 72 hours). By analyzing the fluctuation amplitude, change rate and other characteristics of the data, a traumatic brain injury diagnosis report containing the detection results, data trends and evaluation opinions is generated.

[0068] Example 1: see Figure 2When the fluorescence signal intensity is processed by using the wavelet transform algorithm, a suitable wavelet basis function is selected first. When selecting a suitable wavelet basis function, the waveform characteristics and noise types of the fluorescence signal are combined, the smoothness of the signal, the distribution of abrupt points and the frequency range of the noise are considered, and a wavelet basis function that can effectively separate the specific fluorescence signal and the interference noise is selected to ensure that the effective signal components related to the binding of differentially expressed molecules can be accurately identified and retained in the subsequent decomposition process. The collected fluorescence signal intensity is subjected to multi-scale decomposition. When the collected fluorescence signal intensity is subjected to multi-scale decomposition, the original signal is decomposed into high-frequency components and low-frequency components at different scales. The high-frequency components mainly correspond to the noise and the abrupt part of the signal, and the low-frequency components reflect the overall trend and main characteristics of the signal. Through energy analysis of the components at each scale, the noise components generated by environmental interference, equipment fluctuation and the like are removed, and the effective signal components containing the binding information of differentially expressed molecules are retained. In the decomposition process, the signal is decomposed into different frequency components. Through layer-by-layer analysis of the energy distribution of each frequency component, irrelevant signal components generated by detection environmental interference, equipment noise and the like are identified and removed, and effective signals related to the binding of differentially expressed molecules are retained, the noise reduction processing of the original fluorescence signal is completed, and the noise-reduced signal is obtained. Subsequently, the principal component analysis method is used to perform feature dimension reduction processing on the noise-reduced signal. The covariance structure of the signal data is calculated, a few comprehensive variables that can reflect the main characteristics of the signal are extracted, the covariance matrix reflecting the correlation between the characteristics of each signal is obtained by calculating the covariance structure of the signal data. The eigenvalues of the matrix reflect the size of the data variation in each direction, and the eigenvectors represent the direction of variation. The linear combination corresponding to the first few eigenvectors with larger eigenvalues is extracted as a comprehensive variable. These comprehensive variables can reflect the main information of the original signal data and are independent of each other, thereby realizing effective extraction and dimension reduction processing of the key characteristics of the signal. These variables constitute the key signal characteristics, thereby simplifying the subsequent data processing process.

[0069] Through the above key signal characteristics, the peak intensity, half-peak width and integral area of the fluorescence signal are further calculated. The peak intensity is the highest value in the fluorescence signal curve, directly reflecting the signal intensity after the detection probe binds with the differentially expressed molecules; the half-peak width is the signal width corresponding to the signal intensity reaching half of the peak value, which can reflect the stability of the signal; the integral area is the area surrounded by the fluorescence signal curve and the baseline, which can comprehensively reflect the total amount of the signal. These calculated parameters are compared with the reference values (i.e. preset threshold values) in the bioinformatics database one by one. The reference values are derived from the statistical results of the detection data of a large number of healthy people and known traumatic brain injury patients. The deviation proportion of each parameter from the corresponding reference value is calculated, and the weight of each parameter in the detection is combined. The weight of each parameter is determined by combining its biological significance and clinical relevance in the detection of traumatic brain injury. By analyzing the detection data of a large number of healthy people and traumatic brain injury patients in the bioinformatics database, the correlation degree of parameters such as peak intensity, half-peak width and integral area with injury degree is statistically analyzed. Parameters with high correlation degree are given high weight, and parameters with low correlation degree are given low weight. At the same time, the evaluation opinions of clinical experts on the diagnostic value of each parameter can be referred to. Finally, the weight distribution scheme that can accurately reflect the signal deviation degree is determined, and the comprehensive signal deviation degree is obtained. The deviation degree is used to quantify the difference degree between the current detection signal and the normal reference range.

[0070] In the classification of biological samples, the grid search method is used to optimize the kernel function parameters of the support vector machine classification model. During optimization, the candidate value range of related parameters (such as gamma and penalty coefficient C) is set according to the kernel function type (such as radial basis function), and all possible parameter combinations are generated within the range with a certain step size. Each set of parameters is substituted into the model, and the classification effect of the model on known samples is evaluated by combining cross-validation. The parameter combination that is comprehensive optimal in terms of classification accuracy, sensitivity and specificity is selected as the optimal parameter of the support vector machine classification model. By setting the value range of the parameters, different parameter combinations are traversed within the range, and after substituting each set of parameters into the model, the classification effect of the model on known samples is evaluated by combining cross-validation. In the evaluation of the classification effect of each set of parameters after being substituted into the model, the known samples are divided into training set and test set by cross-validation. The accuracy, sensitivity and specificity are calculated by the classification results of the test set by the model. Among them, the accuracy is the proportion of correctly classified samples to the total number of test samples, the sensitivity is the proportion of correctly classified positive samples to the actual positive samples, and the specificity is the proportion of correctly classified negative samples to the actual negative samples. By comparing the values of the three indicators, the parameter combination with high accuracy, sensitivity and specificity is selected as the optimal parameter combination. The parameter combination with the best classification effect is selected and determined as the optimal parameter of the support vector machine classification model. The signal deviation obtained before is input into the support vector machine classification model optimized by parameters, and the model classifies the biological samples into different categories according to the size and distribution characteristics of the signal deviation, and outputs the preliminary classification results, which usually include positive, negative and suspected.

[0071] Based on the above preliminary classification results, the performance of the support vector machine classification model is evaluated by K-fold cross-validation method. All detection data are randomly divided into K subsets of approximately equal size. Each time, K-1 subsets are selected as the training set to train the model, and the remaining 1 subset is used as the test set to verify the classification effect of the model. This process is repeated K times so that each subset has the opportunity to be the test set. The accuracy, sensitivity and specificity indicators of each validation are calculated, and the average value is taken as the final performance indicator of the model. The accuracy is the proportion of correctly classified samples to the total number of samples, the sensitivity is the proportion of correctly identified positive samples to the actual positive samples, and the specificity is the proportion of correctly identified negative samples to the actual negative samples.

[0072] According to the obtained accuracy, sensitivity and specificity indicators, the decision boundary of the support vector machine classification model is adjusted. The adjustment of the decision boundary is based on the performance of the model in different indicators. If the sensitivity is low, the judgment standard for positive samples is appropriately relaxed, and the judgment threshold of positive samples is reduced, that is, the critical value of the standardized signal value for distinguishing positive and negative is reduced. For example, the original judgment threshold is a fixed value, and after relaxation, the value is lowered, so that more samples with signal deviation exceeding the lowered threshold are judged as positive, thereby increasing the number of correctly identified actual positive samples, and improving the sensitivity of the support vector machine classification model. If the specificity is insufficient, the judgment threshold for positive samples is appropriately increased. Through multiple fine-tuning of the decision boundary, the signal deviation is re-classified to obtain the optimized classification result. The preliminary detection result of traumatic brain injury is integrated by combining the preliminary classification result, the model performance indicators and the optimized classification result.

[0073] Using the above preliminary detection result, the patient's clinical symptom description is compared and analyzed. The clinical symptom description includes the patient's trauma history, consciousness state, headache degree, vomiting condition and other information. The reliability of the biological sample detection is verified by checking the consistency of the preliminary detection result and these clinical symptoms. If the preliminary detection result is highly consistent with the clinical symptoms, it indicates that the detection reliability is high; if there is obvious contradiction, the cause of the contradiction needs to be analyzed, which may be due to improper sample collection, interference factors in the detection process or model parameter setting problems, etc. According to the verification result, a verification report is generated, which records the consistency analysis process and possible problems in detail. Based on the verification report, the preset threshold is adjusted. If multiple verifications show that the current threshold leads to more misjudgments, the threshold value is adjusted upwards or downwards. After adjustment, the comparison result of the standardized signal value and the new threshold is recalculated, and then the preliminary detection result is updated to improve the accuracy and reliability of the detection result.

[0074] Embodiment 2: see Figure 3, the evaluation function is constructed according to the accuracy, sensitivity and specificity indexes, the evaluation function comprehensively considers the values of the three indexes, and the influence of each index in classification is balanced by giving different weights. In the construction process, the actual role of the index in the detection of traumatic brain injury needs to be referred to, and the values of each index are integrated into a comprehensive score in proportion. The actual role of the index is measured by its clinical relevance in the diagnosis of traumatic brain injury, and is specifically combined with a large amount of clinical case data statistical analysis: the accuracy reflects the correctness of the overall classification of the model, and its role is reflected in the reliability of the comprehensive judgment of the detection result; the sensitivity reflects the ability of the model to identify actual positive samples, and its role is to reduce the risk of missed diagnosis, which is particularly important for avoiding missed diagnosis of mild injury; the specificity reflects the ability of the model to distinguish negative samples, and its role is to reduce misdiagnosis, which is crucial for excluding non-injury samples. The actual role of each index is determined by analyzing the performance of different indexes in confirmed cases, such as indexes with high sensitivity play a more significant role in early detection of mild injury. The goal is to maximize the score of the evaluation function, and the gradient descent method is used to adjust the decision boundary parameters of the support vector machine classification model. In the adjustment process, the initial decision boundary parameter value is first set, then the score of the evaluation function under the current parameter is calculated, the parameter value is changed constantly and the score change direction is observed, and the parameter is adjusted gradually to approach the optimal value. The decision boundary parameters include the kernel function parameters of the support vector machine model (such as the bandwidth parameter of the radial basis function), the penalty coefficient, etc. During adjustment, the score of the evaluation function under the current parameter is recorded first, if the score increases with the increase of a certain parameter, the parameter is increased by a certain step; if the score decreases with the increase of the parameter, the parameter is decreased; for multiple parameter combinations, the parameters are adjusted one by one and the score change is observed, and the parameter combination that makes the score of the evaluation function maximum is gradually approached, until the score no longer obviously increases, and the final decision boundary parameter is determined. The adjustment range of each parameter is determined according to the change rate of the current score, to avoid the parameter deviating from the optimal range due to the adjustment range being too large, or prolonging the optimization process due to the adjustment range being too small, until the score of the evaluation function no longer obviously increases, and the parameter adjustment is stopped.

[0075] Based on the adjusted decision boundary parameters, the signal deviation is re-input into the support vector machine classification model for processing. The model analyzes the characteristics of the biological sample represented by the signal deviation according to the new decision boundary, divides the sample into the corresponding category, and generates an intermediate classification result. The intermediate classification result includes the classification label of each sample, such as positive, negative or suspected, and records the distance of each sample from the decision boundary during classification, and the farther the distance, the higher the certainty of classification.

[0076] The consistency of the intermediate classification result and the preliminary classification result is verified, and the labels of each sample in the two classification results are compared one by one during the verification process. For the samples with consistent labels, the number and the proportion in the total samples are counted; for the samples with inconsistent labels, the specific information and the type of label difference, such as from positive to negative, from suspected to positive, etc. are recorded. The Kappa coefficient is a statistical quantity used to quantify the consistency between the intermediate classification result and the preliminary classification result, and its core function is to exclude the interference of random factors on the consistency of classification, so as to more objectively reflect the actual coincidence level of the two classification results. The coefficient measures the reliability of the classification result by comparing the actual observed consistency rate with the expected consistency rate under random conditions, and its value range is [-1, 1]. The closer the value is to 1, the higher the consistency of the two classification results; the closer the value is to -1, the lower the consistency; and when the value is 0, the consistency is equivalent to the random condition. When calculating the Kappa coefficient, first count the number of samples that are actually consistent in the intermediate classification result and the preliminary classification result to obtain the actual consistency rate; then calculate the expected consistency rate under random conditions according to the proportion of positive, negative and suspected samples in the two results; specifically, for the three categories of positive, negative and suspected, the product of the proportion of samples in the preliminary classification result and the proportion of samples in the intermediate classification result is calculated for each category, and the sum of the three products is added to obtain the expected consistency rate under random conditions, which reflects the probability of the coincidence of the two classification results caused purely by accidental factors. Subtract the expected consistency rate from the actual consistency rate, and then divide by 1 minus the expected consistency rate to obtain the Kappa coefficient, which excludes the interference of random factors on the consistency and truly reflects the coincidence degree of the two results.

[0077] If the calculated Kappa coefficient is greater than the preset threshold, it indicates that the intermediate classification result and the preliminary classification result have high consistency, and the intermediate classification result can be determined as the optimized classification result. The preliminary detection result is obtained through signal processing, support vector machine preliminary classification and cross-validation, and has a certain reliability; the intermediate classification result is generated by optimizing the decision boundary parameter based on the preliminary result, and its high consistency with the preliminary result indicates that the optimization process does not deviate from the reasonable classification logic, and the intermediate result is processed by maximizing the evaluation function, and the comprehensive performance is better. Even if the preliminary result has slight deviation, the high consistency indicates that the intermediate result has corrected the deviation, so it can be determined as the optimized result. At this time, the parameter adjustment record in the classification process, the consistency verification details and the final classification label need to be sorted out to form a complete classification report, and the classification basis and the certainty degree of each sample need to be clearly marked in the report.

[0078] If the Kappa coefficient is less than or equal to the preset threshold, it indicates that the consistency of the two classification results is insufficient, and the reliability of the intermediate classification result needs to be verified. At this time, the adjustment process of the decision boundary parameter needs to be reexamined, and whether the parameter adjustment direction is reasonable and whether there is adjustment deviation caused by local optimal value needs to be analyzed. The range and step of parameter adjustment are reset, and the gradient descent method is used again to adjust the decision boundary parameter. After adjustment, the signal deviation degree is classified and processed again to generate a new intermediate classification result. When resetting the parameter adjustment range, refer to the historical adjustment record. If the previous range does not cover the area with a higher score of the evaluation function, the range is expanded (for example, the original range is 0.1-1.0, which can be adjusted to 0.05-2.0). The step size is set according to the parameter sensitivity. If the parameter fine-tuning leads to a significant change in the score, the step size is reduced (for example, from 0.1 to 0.05), otherwise the step size is increased. During adjustment, the parameters are changed in the new range and step size, the score of the evaluation function is calculated, the parameter direction that increases the score is retained, and the better parameter combination is found. Repeat the consistency test and calculate the Kappa coefficient until the Kappa coefficient obtained is greater than the preset threshold. The intermediate classification result at this time is taken as the optimized classification result. In this process, the specific values of each parameter adjustment, the changes of the classification result and the fluctuations of the Kappa coefficient need to be recorded to trace the key nodes in the adjustment process and provide a reference for subsequent model optimization. At the same time, for samples that are still difficult to reach the preset consistency after multiple adjustments, the signal deviation degree characteristics of the samples need to be analyzed to determine whether the classification is difficult due to the particularity of the samples themselves or whether there is an abnormal situation that the model cannot cover. For the separately marked samples, the specific values of the signal deviation degree, the fluctuation amplitude and the deviation mode from the reference value in the bioinformatics database are extracted, and the signal deviation degree characteristic distribution of the healthy population and the typical traumatic brain injury patients is compared. If the sample deviation value is far beyond the normal range or the fluctuation mode does not match the known pathological characteristics, it is judged as the particularity of the sample itself (such as the sample being affected by a special physiological state); if its characteristics are significantly different from all category characteristics in the model training data and cannot be covered by the existing classification logic, it is judged as an abnormal situation that the model does not cover. Provide direction for further improvement of the classification model.

[0079] Example 3: see Figure 4, based on the differentially expressed molecules and the normalized signal values, a time series data matrix is constructed, which needs to cover all the detected differentially expressed molecules, concentrations and corresponding normalized signal values, and is arranged in order of detection time. The row dimension of the matrix corresponds to different detection indicators, including the concentration data of each differentially expressed molecule, the normalized signal values obtained by each detection, etc.; the column dimension corresponds to different detection time points, and the interval of the time points is determined according to the detection frequency, which can be several hours, one day or several days. During the construction process, the integrity of the data needs to be ensured, and for the missing detection data, an interpolation method is used for supplement, and when interpolating, the numerical change trend of the adjacent time points is referred to, so that the supplemented data maintains the continuous change characteristics.

[0080] The time series data matrix constructed is divided into a sliding window, and the size of the window is set according to the detection period and the data fluctuation characteristics, for example, when the detection time point interval is 12 hours and the change trend within 3 days needs to be captured, the window size can be set to data containing 6 time points. The sliding step of the window is set to 1 time point, that is, each subsequent window moves one time point backward on the basis of the previous window, and in this way, multiple continuous and overlapping sub-matrices can be obtained, each of which represents the detection data in a specific time period.

[0081] The mean, variance and autocorrelation coefficient in each time window are extracted as dynamic feature vectors. The calculation of the mean is the sum of all data in the window divided by the number of data, which reflects the average level of the detection indicators in the time period; the calculation of the variance is the sum of the square of the difference between each data and the mean divided by the number of data, which reflects the dispersion degree of the data; the calculation of the autocorrelation coefficient needs to set the lag order, for example, when the lag order is 1, the correlation degree between the data in the window and the data after moving 1 position backward is calculated, and the formula is as follows:

[0082]

[0083] wherein, represents the autocorrelation coefficient, represents the i-th data in the window, represents the mean of the data in the window, represents the number of data in the window. The calculated mean, variance and autocorrelation coefficient are combined in a fixed order to form the dynamic feature vector corresponding to each window, and the dimension of the feature vector is 3 times the number of detection indicators, that is, each detection indicator corresponds to three characteristic values of mean, variance and autocorrelation coefficient.

[0084] The dynamic feature vector is input into a pre-trained long short-term memory network model, which has been trained through a large amount of historical detection data. The model structure includes an input layer, multiple hidden layers, and an output layer. The hidden layers contain structures such as forget gates, input gates, and output gates for processing long-term dependencies in time series data. The input layer of the model receives the dynamic feature vector and converts it into a tensor form suitable for network processing. The hidden layers process the input feature vector through a gating mechanism, selectively retaining or forgetting historical information, and updating the network state according to the current input. The output layer outputs the trend prediction value based on the final state of the hidden layer. The prediction value is a vector containing the predicted values of each detection indicator at one or more future time points.

[0085] According to the deviation of the trend prediction value from the historical detection data, the prediction error at each time point is calculated as the absolute value of the difference between the prediction value and the actual value at the corresponding time point in the historical detection data. The actual detection data, trend prediction value, and prediction error are presented in the form of a curve graph using a data visualization tool to draw a trend analysis graph. The horizontal axis of the graph is time, and the vertical axis is the value of the detection indicator. The actual data curve is represented by a solid line, the prediction data curve is represented by a dashed line, and the error range is shown by the shaded area around the curve.

[0086] The trend analysis graph is integrated with detailed information of the differentially expressed molecules (including the name of the molecule, biological function, known role in traumatic brain injury, etc.), specific values of standardized signal values, preliminary detection results, etc. to generate a diagnostic report for traumatic brain injury. The structure of the report needs to be clear and concise. First, list the basic information of the detection, including the detection object, detection time, detection indicators, etc. Then present the types and content changes of differentially expressed molecules. Then show the standardized signal values and preliminary detection results. Then show the trend and prediction of the detection data through the trend analysis graph. Finally, based on all the above information, give a comprehensive diagnostic conclusion. The conclusion needs to explain the possible degree and trend of traumatic brain injury reflected by the detection data, and label the possible interference factors and uncertainties of the data in the detection process.

[0087] Example 4: see Figure 5In the training phase of the long short-term memory network model, the hidden layer weights are updated using the time-reversal propagation algorithm. The historical dynamic feature vectors are input into the model in chronological order, which come from the detection data of past traumatic brain injury patients and contain mean, variance, and autocorrelation coefficients within different time windows. When the model processes these vectors, it calculates the difference between the output result and the actual detection result, which is the loss value. Starting from the output layer, this loss value is passed back along the time axis, adjusting the connection weights between neurons in the hidden layer layer by layer. Each adjustment is based on the direction of change in the loss value, allowing the model to reduce bias when processing similar data in the future. This process is repeated until the model's processing results for historical data stabilize within a certain range.

[0088] The model controls the forgetting and remembering ratio of feature vectors through a gating mechanism. The forget gate determines whether to retain the previously stored cell state information based on the current input dynamic feature vector and the hidden state at the previous time. For example, when the newly input feature vector shows a significant change in the detection indicator, the forget gate will reduce the retention ratio of the old information so that the model can focus more on the latest data. The input gate is responsible for filtering key information in the current feature vector and storing it in the cell state. For example, when the concentration of a certain differentially expressed molecule suddenly increases, the input gate will increase the memory weight of this information. The output gate determines the information output to the next time based on the current cell state and hidden state, ensuring that the model can continuously track the trend of the detection data.

[0089] A dropout layer is set after the fully connected layer of the model. When the dynamic feature vector passes through the fully connected layer, the dropout layer will temporarily disable a portion of the neurons at random, and the proportion of disabled neurons is set according to the model training conditions. The model training conditions refer to the training conditions of the long short-term memory network model, including the degree of overfitting during training, the change in prediction accuracy on the validation set, and the fluctuation of the loss value. When overfitting occurs during model training (the training accuracy is much higher than the validation accuracy), the proportion of disabled neurons can be increased; if the model is underfit, the proportion of disabled neurons can be decreased. The fully connected layer is a level in the long short-term memory network located after the hidden layer, where each neuron is connected to all neurons in the previous layer to integrate the feature information output by the hidden layer and convert it into comprehensive features suitable for processing by the output layer, providing a basis for the output of subsequent trend prediction values. For example, if the proportion is set to 0.3, approximately 30% of the neurons do not participate in calculation during each training. This approach can prevent neurons from relying too much on specific input features, making the model more versatile when processing new detection data. After the dropout layer, the model output layer converts the result into a normalized trend prediction value between 0 and 1, facilitating the comparison of trends for different detection indicators.

[0090] When calculating the moving average of the normalized signal value in continuous multiple detections, the number of calculations needs to be determined, such as selecting the latest 5 detection results. Each time the calculation is performed, the latest 5 normalized signal values are taken, and the arithmetic mean is taken as the current moving average. For example, the normalized signal values of the first to fifth times are a, b, c, d, and e, and the moving average is (a+b+c+d+e) / 5; when the sixth detection is completed, the moving average is updated to (b+c+d+e+f) / 5, where f is the normalized signal value of the sixth time.

[0091] According to the matching degree of the moving average and the verification report result, the floating range of the preset threshold is dynamically adjusted. The matching degree of the verification report result refers to the average level of the moving average of the normalized signal value in continuous multiple detections, and the verification report result is the verification conclusion of the reliability of the detection combined with the clinical symptoms. The matching degree of the two is the degree of coincidence of the detection result corresponding to the moving average and the verification conclusion. If the matching degree is high, it means that the current threshold can effectively distinguish between normal and abnormal signals, and the floating range can be reduced to improve the detection specificity; if the matching degree is low, it means that the current threshold may miss abnormal signals, and the floating range needs to be expanded to ensure that potential abnormal conditions are covered, thereby adapting to the needs of different detection scenarios. If the detection result corresponding to the moving average is consistent with the majority of the clinical diagnosis in the verification report, it means that the current threshold is reasonable, and the floating range can be reduced, such as from ±10% to ±5%. If the result corresponding to the moving average does not match the verification report multiple times, the floating range of the threshold needs to be expanded, such as to ±15%, so that more possible abnormal signals are included in the detection range.

[0092] When the matching degree is lower than the set standard, the re-calibration process of the quantum dot labeled detection probe is triggered. The standard distribution curve of the historical fluorescence signal intensity is obtained, which is obtained by detecting the standard sample multiple times and presents the typical normal distribution characteristics. The K-means clustering algorithm is used to process the current fluorescence signal intensity, and the signal data is divided into two categories: normal cluster and outlier cluster. For example, if the detected fluorescence signal intensity is mostly concentrated in 100-300 units, and a small number of signals are below 50 or above 500 units, the algorithm will mark these signals deviating from the main range as outliers. The proportion of outliers in the total signal is calculated, and if the proportion exceeds the warning value, such as reaching 20%, the coupling reaction parameter optimization program of the recombinant antibody and quantum dots is started. The pH value in the reaction system is adjusted, such as from the original 7.4 to 7.2; the reaction time is extended, from 2 hours to 2.5 hours; or the mixing ratio of the recombinant antibody and quantum dots is changed, from 1:20 to 1:15. Through the optimization of these parameters, the detection probe is prepared again to improve the stability and specificity of its combination with the differentially expressed molecules.

[0093] In the process of establishing a multi-round iteration record library for a support vector machine classification model, key information in each iteration of the model needs to be collected. These include the number of iterations, the type of kernel function used, the specific values of the kernel function parameters, the coordinate parameters of the decision boundary, the input signal deviation dataset, the output classification result label, and the accuracy, sensitivity, and specificity values of the model in that iteration. The record library can be stored in the form of a structured data table, with column headers corresponding to the above information, and each row corresponding to the record of an iteration process. For example, in the first iteration, the kernel function is a radial basis function, the parameter gamma takes the value 0.1, the penalty coefficient C takes the value 10, the decision boundary parameters are a set of specific values, the input signal deviation contains data from 50 samples, the classification results include 30 positives, 15 negatives, and 5 suspected, the accuracy is 0.85, the sensitivity is 0.82, and the specificity is 0.88. All of this information needs to be completely recorded in the record library.

[0094] When extracting the decision boundary parameter variation and classification result difference in each iteration, the records of adjacent two iterations need to be compared. The decision boundary parameter variation is the difference between the parameter values of the latter iteration and the former iteration. If the parameter is multi-dimensional, the variation of each dimension is calculated separately. For example, the decision boundary parameters of the first iteration are (2.3, 4.1), and the parameters of the second iteration are (2.5, 3.9), then the variation is (0.2, -0.2). The classification result difference includes the increase or decrease of positive sample number, the transformation between negative samples and suspected samples, etc. For example, there are 30 positive samples in the first iteration, and 28 positive samples in the second iteration, of which 2 are converted to suspected, then record that the number of positive samples decreases by 2 and 2 are converted to suspected. Arrange these variations and differences in sequence according to the iteration order, and each sequence corresponds to the variation of an index.

[0095] When recognizing the parameter variation pattern through a convolutional neural network, the arranged parameter variation sequence and classification result difference sequence need to be converted into a format suitable for network input. For example, each sequence is divided into multiple segments of fixed length, and each segment is used as the input sample of the network. The input layer of the convolutional neural network receives these samples, and through the first convolutional layer, multiple different size convolution kernels are used to slide and calculate the samples, extracting local parameter variation features such as the increasing or decreasing trend of parameters in consecutive iterations. After the convolutional layer processing, the feature map is compressed through the pooling layer, retaining key features while reducing data volume. Subsequently, these features are transmitted to multiple hidden layers for further processing, gradually integrating local features to form a representation of the overall parameter variation pattern. Finally, the output layer of the network outputs a model stability evaluation index, which is a specific numerical value reflecting the stability of the model under the current parameter variation pattern.

[0096] When the evaluation index is spatiotemporally correlated with the signal deviation, the corresponding relationship between the two in time and value needs to be considered. For example, the evaluation index of each iteration is paired with the signal deviation data used in the corresponding iteration in the order of iterations to form multiple sets of correlated data. Statistical analysis is performed on these data. When the evaluation index is at a high level, the numerical range, concentrated distribution interval, and dispersion degree of the statistical signal deviation are summarized to summarize its distribution pattern. When the evaluation index is at a low level, the change trend of the signal deviation with the decrease of the evaluation index is observed, including the numerical increase and decrease amplitude, fluctuation frequency, and whether there is an abnormal jump. By comparing the performance of the signal deviation in different evaluation index intervals, the correlation between the two is determined to provide a basis for subsequent division of the confidence level of the detection data. According to the mapping results, the confidence level of the detection data is divided, for example, when the evaluation index is greater than 0.8 and the signal deviation is greater than 2.0, the corresponding data is divided into a high confidence level; when the evaluation index is less than 0.5 and the signal deviation is less than 1.0, the corresponding data is divided into a low confidence level; and other cases are of medium confidence level.

[0097] When the confidence level is fused with the optimized classification result to generate the final detection report framework, different processing methods need to be adopted according to different confidence levels. For data of high confidence level, the optimized classification result is directly adopted, and the high reliability of the result is indicated in the report; for data of medium confidence level, the classification result is retained while the parameter changes of the data in the iteration process are supplemented for reference; for data of low confidence level, the corresponding clinical symptom description needs to be attached to explain the uncertainty of the classification result. The structure of the report framework includes sample basic information, key parameter summary of each iteration, correlation map of signal deviation and evaluation index, classification result summary table, and explanation of data of different confidence levels. For example, in the classification result summary table, each row corresponds to a sample, and the column titles include sample number, optimized classification result, confidence level, evaluation index, signal deviation, etc., so that the reader can clearly understand the detection situation and reliability of each sample.

[0098] It should be noted that, in this text, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0099] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A rapid detection method of traumatic brain injury by fusing genetic engineering and quantum dots, characterized in that, include: The intensity of the fluorescence signal generated after the quantum dot-labeled detection probe binds to differentially expressed molecules in biological samples is collected using a fluorescence detection device. The fluorescence signal intensity is denoised and its features are extracted using signal processing algorithms to obtain standardized signal values. Based on the comparison results of the standardized signal values ​​with preset thresholds, the biological sample is evaluated in conjunction with a bioinformatics database to determine the preliminary detection results of traumatic brain injury. Based on the differentially expressed molecules, the standardized signal values, and the preliminary detection results, a multivariate statistical model is used to analyze the trend of changes in the detection data within a preset time window, and a diagnostic report of traumatic brain injury is generated. in: The fluorescence signal intensity is decomposed into multiple scales using wavelet transform algorithm to remove noise interference components and obtain a denoised signal. Principal component analysis is then used to perform feature dimensionality reduction on the denoised signal to extract key signal features. The peak intensity, half-width at half-maximum (WHM), and integral area of ​​the fluorescence signal are calculated using the key signal features. The peak intensity, WHM, and integral area are then compared with preset thresholds in the bioinformatics database to obtain the signal deviation. Based on the signal deviation, the biological samples are classified using a support vector machine classification model, the classification results are output, and the classification results are optimized using cross-validation to obtain preliminary detection results of traumatic brain injury. Using the preliminary test results and combined with the description of clinical symptoms, the reliability of the biological sample detection is verified, a verification report is generated, and the preset threshold is adjusted and the preliminary test results are updated based on the verification report. in: The kernel function parameters of the support vector machine classification model are optimized using the grid search method to determine the optimal parameter combination. The signal deviation is then input into the support vector machine classification model to generate preliminary classification results. Based on the preliminary classification results, the performance of the support vector machine classification model is evaluated using K-fold cross-validation, and the accuracy, sensitivity, and specificity indices are calculated. Based on the accuracy, sensitivity, and specificity indices, the decision boundary of the support vector machine classification model is adjusted, and the signal deviation is reclassified to obtain the optimized classification result. By combining the preliminary classification results, the accuracy, sensitivity, and specificity indices, and the optimized classification results, preliminary detection results for traumatic brain injury are generated. in: A time series data matrix is ​​constructed based on the differentially expressed molecules and standardized signal values, and the time series data matrix is ​​then segmented using a sliding window. The mean, variance, and autocorrelation coefficient within each time window are extracted as dynamic feature vectors. The dynamic feature vector is input into a pre-trained long short-term memory network model, and the trend prediction value is output. Based on the deviation between the predicted trend value and historical detection data, a trend analysis graph is generated and integrated into the diagnostic report.

2. The fusion gene engineered and quantum dot-based rapid detection method of traumatic brain injury according to claim 1, characterized in that, The step involves adjusting the decision boundary of the support vector machine classification model based on the accuracy, sensitivity, and specificity indices, and then reclassifying the signal deviation to obtain an optimized classification result, including: An evaluation function is constructed using the accuracy, sensitivity and specificity indicators, and a decision boundary parameter of the support vector machine classification model is adjusted by a gradient descent method to maximize the evaluation function; The signal deviation is reclassified based on the adjusted decision boundary parameter to generate an intermediate classification result; The intermediate classification result and the preliminary classification result are subjected to consistency test, a Kappa coefficient is calculated, and the classification consistency is determined according to the Kappa coefficient; If the Kappa coefficient is greater than a preset threshold, the intermediate classification result is taken as an optimized classification result; if the Kappa coefficient is less than or equal to the preset threshold, the decision boundary parameter is repeatedly adjusted and the classification is re-performed until the optimized classification result is obtained.

3. The fusion gene engineered and quantum dot-based rapid detection method of traumatic brain injury according to claim 1, characterized in that, The dynamic feature vector is input into a pre-trained long short-term memory network model, and a trend prediction value is output, including: The time back propagation algorithm is used to update the hidden layer weight of the long short-term memory network model; The forgetting and memory ratio of the feature vector is controlled through a gating mechanism; The dropout layer is used to regularize the full connection layer, and a normalized trend prediction value is output.

4. The fusion gene engineered and quantum dot-based rapid detection method of traumatic brain injury according to claim 1, characterized in that, The preset threshold is adjusted based on the verification report, including: The moving average of the normalized signal value in continuous multiple detections is calculated; The floating range of the preset threshold is dynamically adjusted according to the matching degree of the moving average and the verification report; When the matching degree is lower than the set standard, the re-calibration process of the quantum dot labeled detection probe is triggered.

5. The fusion gene engineered and quantum dot-based rapid detection method of traumatic brain injury according to claim 4, characterized in that, The re-calibration process of the quantum dot labeled detection probe includes: The standard distribution curve of the historical fluorescence signal intensity is obtained, and the K-means clustering algorithm is used to detect outliers of the current fluorescence signal intensity; If the proportion of outliers exceeds the warning value, the coupling reaction parameter optimization program of the recombinant antibody and the quantum dot is started.

6. The fusion gene engineered and quantum dot-based rapid detection method of traumatic brain injury according to claim 1, characterized in that, After the preliminary detection result of the traumatic brain injury is generated, the method further includes: A multi-round iteration record library of the support vector machine classification model is established; The decision boundary parameter change amount and the classification result difference of each iteration are extracted; The parameter change mode is recognized by a convolutional neural network to generate a model stability evaluation index.

7. The fusion gene engineered and quantum dot-based rapid detection method of traumatic brain injury according to claim 6, characterized in that, The method further includes: The model stability evaluation index and the signal deviation are subjected to spatio-temporal correlation mapping, the confidence level of the detection data is divided according to the mapping result, and a final detection report framework is generated by fusing the confidence level and the optimized classification result.

Images